GPT-6 Astra Hits 39% on OpenDS-World Live Web Deep Search Benchmark vs 21.1% Human Average
emmanuelvivier · x · 2026-10-25
The arXiv paper "Deep Search in Real-World Web Environments" introduces OpenDS and the OpenDS-World benchmark: 200 challenging tasks requiring agents to autonomously perform multi-step deep search on the live open Web, discovering and integrating evidence across independent sources via web search, browser interaction, and evidence processing to reach a unique verified factual answer. Evaluating 12 frontier models under a unified tool environment, the best performer — reported as GPT-6 Astra — scores 39%, well above the 21.1% human average.
More from Models
- Pro user: 62,500 Codex credits ($2,500) lock to smallest models the day your plan ends — lanqo88 · 2026-10-25
- Could OpenAI throttle Codex usage differently per user? Redditors demand transparency — Dadideology · 2026-10-25
- Claude notices it's Sunday for a European user and tells them to go touch grass — aiamblichus · 2026-10-25
- Claude Models Now GA on Google Cloud with New Anthropic-Managed Deployment Option — emmanuelvivier · 2026-10-25
- Study: Grok picks Republicans 92%, GPT-6.1 Sol picks Democrats 97% — emmanuelvivier · 2026-10-25
- Microsoft's MAI-DxO solves 83.2% of complex cases vs physicians' 36.1% — emmanuelvivier · 2026-10-25